The heads of OpenAI, Anthropic, Google DeepMind, and Microsoft AI signed a joint open letter to US Congress on June 5, calling for federal legislation requiring all US synthetic DNA providers to screen orders for potentially dangerous genetic sequences (Fortune). Signatories include Sam Altman, Dario Amodei, Demis Hassabis, and Mustafa Suleyman, alongside Nobel laureate scientists and former US defense officials (Interesting Engineering). The letter argues that AI has substantially lowered the technical expertise threshold previously required to design and weaponize biological material, and that synthetic DNA and RNA can currently be ordered online with inconsistent or absent screening (Microsoft On the Issues). The signatories back the Biosecurity Modernization and Innovation Act of 2026 - introduced by Senators Tom Cotton and Amy Klobuchar - which would mandate both biosecurity screening and order record-keeping from DNA synthesis vendors (Interesting Engineering). A concurrent Nature analysis noted that AI tools for protein and genome design have reached a point where capabilities previously requiring specialized wet-lab training are becoming broadly accessible, compressing a technical barrier that biosecurity frameworks had historically relied on (Nature).
Separately, researchers from the University of Toronto, the Vector Institute, and the University of Cambridge presented a proof-of-concept agentic worm at Infosecurity Europe in early June that uses a free open-weight large language model to reason about each target and compose attacks on the fly, without relying on fixed exploit code (The Register). Across 15 fully autonomous seven-day runs against a simulated enterprise network, the worm correctly identified approximately 82 percent of vulnerabilities it eventually exploited, compromised around 74 percent of network hosts on average, and achieved up to seven generations of self-replication without human assistance (Fortune). The system can ingest newly published vulnerability advisories at runtime, converting recently disclosed CVEs into functional exploits in days rather than the months traditional malware campaigns require (Help Net Security). A Cisco survey cited at the conference found that only 29 percent of organizations reported readiness to defend agentic AI deployments, while 83 percent planned to move them into production business functions (TechTimes).